Add specifications for two-times upscale and unified depth plane features
- Introduced the two-times upscale specification detailing requirements for unified output, fixed model presets, arbitrary dimensions, and consistent job behavior. - Added scenarios for existing models, repeated upscale, and handling of alpha preservation. - Created the unified depth plane specification outlining geometry modes, stable interface, and asset evaluation. - Defined scenarios for mode switching, relief and camera preservation, and validation of delivered assets.
This commit is contained in:
68
openspec/specs/compact-repeat-arrangement/spec.md
Normal file
68
openspec/specs/compact-repeat-arrangement/spec.md
Normal file
@@ -0,0 +1,68 @@
|
||||
# compact-repeat-arrangement Specification
|
||||
|
||||
## Purpose
|
||||
TBD - created by archiving change compact-repeat-node-arrangement. Update Purpose after archive.
|
||||
## Requirements
|
||||
### Requirement: Compact repeat execution path
|
||||
|
||||
排列器 SHALL 按循环执行的几何步骤安排 Repeat 两端,取值支线独立排列。
|
||||
|
||||
#### Scenario: Set Position loop with long value calculations
|
||||
- **WHEN** 循环只执行 Set Position,输入值来自长计算链
|
||||
- **THEN** Repeat 两端保持围绕 Set Position 的紧凑排列,预览执行区域按几何主链确定
|
||||
|
||||
### Requirement: Geometry processing order
|
||||
|
||||
排列器 SHALL 将前置循环安排在后续几何处理之前。
|
||||
|
||||
#### Scenario: Depth Cutout boundary smoothing
|
||||
- **WHEN** 排列 O Image Depth Cutout 的前置边界平滑循环
|
||||
- **THEN** 循环位于其后续几何消费者左侧,保持原有连接与求值结果
|
||||
|
||||
### Requirement: Readable field branches and geometry references
|
||||
|
||||
排列器 SHALL 将取值支线紧凑安排在消费者附近,复杂支线使用互不重叠的 Frame,并分离几何引用的水平通道。
|
||||
|
||||
#### Scenario: Long calculation chain
|
||||
- **WHEN** 消费者使用较长的简单计算链
|
||||
- **THEN** 计算链使用紧凑折叠及分段纵向布局
|
||||
- **AND** 连续依赖按计算方向展开,独立计算按实际消费输入分组
|
||||
|
||||
#### Scenario: Complex depth cutout branches
|
||||
- **WHEN** 排列 O Image Depth Cutout
|
||||
- **THEN** 复杂计算支线分别成框,几何引用线保持独立的水平通道,并通过斜向曲线过渡,原有几何求值保持一致
|
||||
|
||||
#### Scenario: Geometry trunk alongside its consumer frame
|
||||
- **WHEN** 几何主干向 Frame 内的计算节点供值
|
||||
- **THEN** 主干与完整 Frame 边界保持间距,通过斜向分支接入框内消费者
|
||||
|
||||
#### Scenario: Long branch near its geometry consumer
|
||||
- **WHEN** 较大支线需要占用来源和消费者之间的横向空间
|
||||
- **THEN** 主线在消费者之前预留支线及斜向出口的空间,连续主线保持直接向前连接
|
||||
|
||||
#### Scenario: Independent calculation before a geometry consumer
|
||||
- **WHEN** 支线独立于上游几何结果
|
||||
- **THEN** 支线可向主线起点之前展开,主线按实际出口位置保持紧凑
|
||||
|
||||
#### Scenario: Return connection and neighboring reference tracks
|
||||
- **WHEN** 几何结果连接左侧消费者
|
||||
- **THEN** 连线直接向消费者方向过渡,各水平通道按最终范围保留间距
|
||||
|
||||
#### Scenario: Geometry fanout without overlapping wires
|
||||
- **WHEN** 同一几何结果供多个节点消费
|
||||
- **THEN** 近处消费者直接接入,公共水平通道按从左到右分发,每段公共连线只绘制一次
|
||||
- **AND** 水平连线保持向前,短 reroute 间距合并,近乎平行的曲线与主线保持间距
|
||||
|
||||
#### Scenario: Shared results outside calculation branches
|
||||
- **WHEN** 几何或其他类型的结果供多个支线消费
|
||||
- **THEN** 共享结果沿支线外围的公共管道分发,在消费者附近斜向接入
|
||||
- **AND** 原有主线连接保持执行顺序,公共管道与节点、Frame 及其他管道保留间距
|
||||
|
||||
#### Scenario: Geometry transits an unrelated calculation
|
||||
- **WHEN** 几何连接沿途经过其他支线
|
||||
- **THEN** 整段曲线绕开不消费该结果的 Frame,仅允许进入自身来源或消费者所属的 Frame
|
||||
|
||||
#### Scenario: Peripheral producers feed their shared pipes forward
|
||||
- **WHEN** 支线计算的结果通过公共管道供多个消费者使用
|
||||
- **THEN** 共享计算支线与管道一同位于消费者外围,管道入口在结果输出节点右侧
|
||||
|
||||
Reference in New Issue
Block a user